Google has just released the Pixel 10 series, with all four models equipped with the Tensor G5 processor. The Tensor chip traditionally supports various outstanding features unique to Pixel, but it has lagged behind competitors in peak power and efficiency.
Nevertheless, the Tensor G5 is still a milestone processor for the company, as it is the first Tensor chip produced by TSMC rather than long-time partner Samsung. So, how much do we know about this chip?
Cooler and More Cost-Effective Pixel Chip CONTENT
First, the Tensor G5 is manufactured using TSMC’s 3nm process. Google has not disclosed the specific process used, but the shift from Samsung to TSMC is still good news for Pixel users.
TSMC has traditionally been more proficient in chip manufacturing than Samsung. This has made chips produced by TSMC more efficient and powerful than comparable chips produced by Samsung. Google has not detailed the specific efficiency improvements relative to the Tensor G4 manufactured by Samsung.
It is worth mentioning that Google claims the battery life of the Pixel 10 phones is “over 30 hours,” while the Pixel 9 series had a battery life of “over 24 hours.” This is at least partly due to the new phones having slightly larger battery capacities, and we speculate that other components are also more efficient. However, the Tensor G5 is likely to play a significant role in this aspect as well.
Many previous Pixel phones have also suffered from overheating issues, even in less demanding scenarios and tasks. The good news is that Google has stated that it has upgraded “hardware and software thermal controls,” allowing the chipset to run at higher clock frequencies without throttling.
Google specifically told Android Authority that the standard Pixel 10 uses a “graphite thermal solution,” while the Pro model uses a vapor chamber cooling system. It is currently unclear whether the size of this vapor chamber will be larger. We can only wait and see if these measures are effective or if overheating issues still persist.
Connectivity has been another issue plaguing early Pixel phones equipped with Tensor chips. Google seemed to have addressed this issue last year with the new Exynos Modem 5400. So, what new highlights do we have this time?
The company told us: “Building on last year’s new modem, we are further optimizing this experience through software improvements,” indicating that the Exynos Modem 5400 is a continuation from the Pixel 9 series.
Tensor G5 CPU and GPUGoogle, Why So QuietCONTENTCompared to the Tensor G4, the CPU layout of the Tensor G5 has been improved. Google only pointed out in the press release that the CPU consists of one large core, five medium cores, and two small cores. This represents a significant layout change compared to last year’s chip (one large core, three medium cores, and four small cores).
The company did not disclose the specific cores or clock frequencies used. However, Google later confirmed to us that the CPU has been upgraded to “the latest generation of mid-core IP.” This suggests that the Tensor G5 at least uses Cortex-A725 cores. This partially aligns with our previous exclusive leak, which indicated that the Tensor G5 has one Cortex-X4 core, five Cortex-A725 cores, and two Cortex-A520 cores.
So, what does this mean for CPU performance? Google claims that the CPU of this chip is on average 34% faster than the chip in the Pixel 9. The company did not specify whether this figure refers to single-core or multi-core performance, but I expect that, thanks to the updates and a greater number of mid-cores, multi-core performance will be better. However, I do not expect the Tensor G5 to outperform the Snapdragon 8 Elite chipset and Qualcomm’s custom Kryo cores.
The multi-core CPU performance of the Tensor G5 has seen significant improvements, but Google has been surprisingly silent on the GPU front.
What about the GPU? Google did not disclose any specific information about the Tensor G5 GPU during the launch, only stating that it uses “updated” GPU IP. The company also claimed that the Pixel 10 series performed “exceptionally well” in internal tests running many popular games. However, if you are expecting performance or efficiency improvements, these vague statements are not a good sign.
Our previous leak of Tensor G5 GPU information suggests it is equipped with the Imagination DXT-48-1536 GPU, but we need to confirm this through review samples. This would represent a significant departure from the Arm Mali GPU technology that has been used since the first Tensor processor. Google also stated that the Tensor G5 GPU does not support ray tracing.
In terms of game and emulator support, Qualcomm’s Adreno GPUs and Arm’s Mali graphics dominate. Therefore, if your Pixel 10 phone experiences compatibility issues with certain games and applications, do not be surprised.
Artificial Intelligence and Imaging CONTENT
AI has become a major focus for smartphone processors, so what highlights does Google have in this area? The company claims that the TPU performance of the Tensor G5 has improved by up to 60%. Google states that these improvements stem from architectural upgrades, such as more compute blocks and higher frequencies. Due to the ambiguity of AI benchmark tests, it is still difficult to determine how the AI performance of the Tensor G5 compares to competing chips.
Nevertheless, Google states that the Pixel 10 series supports a so-called “Matryoshka transformation model,” referring to Russian nesting dolls. In any case, this architecture allows a smaller, speed-optimized AI model (2 billion “effective” parameters) to be nested within a larger, quality-focused model (4 billion “effective” parameters). The company states that this approach means the phone does not need two separate models, and applications can “dynamically” choose between the full model and the sub-model.
The Tensor G5 features enhanced AI hardware and supports some interesting AI model tricks.
Interestingly, Google explained to us that the sub-model actually has 5 billion parameters, while the full model actually has 8 billion parameters. However, some of these parameters are not loaded into RAM. Instead, the Pixel manufacturer uses layer-wise embedding technology to extract these parameters from flash memory in “tiny increments.”
So, what do these improvements mean for AI performance? Google claims that the latest Gemini Nano model running on the Tensor G5 is 2.6 times faster than the Tensor G4 in the Pixel when performing tasks such as Pixel screenshot and recorder applications, with efficiency being twice that of the Pixel 9’s Tensor G4. The search giant also claims that the token window (the amount of input the model can process at once) has increased from 12,000 on the Tensor G4 to 32,000 on the new chip. Google states that this larger token window is roughly equivalent to 100 screenshots or a month’s worth of emails.
The upgraded AI chip can deliver “over 20 unique experiences” at launch, and Google states that more features will be rolled out in the future. Some noteworthy built-in features include 100x Pro Res zoom, voice translation during calls (presented in a human voice), personal journal applications, Magic Cue, and more.
The image signal processor (ISP) of the Tensor G5 has also been redesigned, working more closely with the digital signal processor. MediaTek and Qualcomm have also taken a similar approach in recent generations, making their ISPs work more closely with AI chips.
Google claims that the chipset supports “advanced” segmentation features that can identify objects and details in scenes. It illustrates how segmentation can be used in portrait photography to identify skin, lips, eyes, etc., resulting in more accurate images. This is not new, as competitors have long provided advanced or even real-time scene segmentation features in photos and videos. Nevertheless, I am still pleased to see these upgraded features in Pixel.
Google’s new ISP has also implemented several video-related improvements, such as reducing blur in low-light videos and enabling 10-bit HDR by default in 1080p and 4K/30fps video recording. Unfortunately, due to Google’s decision to forgo native 8K recording on the phone, you will still need to rely on cloud-based Video Boost features to record 8K videos.Significant Questions Remain About Tensor G5 CONTENT
Due to the lack of some basic details, it is difficult to draw clear conclusions about the Tensor G5. Google has refused to disclose any performance improvements regarding the GPU compared to the Pixel 9 series, suggesting that performance may not have changed or could even be downgraded. We will also need to wait for review samples to see if the switch to TSMC has had any effect on battery life and whether Google has fully resolved the overheating issues.
However, one thing is certain. Google’s Tensor processors have never truly focused on peak performance or benchmark testing. I expect that the Tensor G5 and Pixel 10 series will not change this status quo.Google Tensor G5 CONTENT
Five Reasons Why Tensor G5 Changes the Pixel Landscape
Our vision for Pixel has always been to create a smartphone that features intelligent practical functions, making your life easier and more convenient. At the core of this is Google’s custom-designed mobile processor, Tensor. Five years ago, when we launched Tensor, it pushed the limits of Pixel phones through advanced machine learning technology. Today, we are working closely with the Google Research team to create a new chapter in AI innovation and introduce the largest upgrade to our custom chip to date: Google Tensor G5.
Five Things to Know About the New Tensor G5 Chip
1. Google Tensor G5 is our largest upgrade to date, bringing significant performance improvements.
We have enhanced the Tensor G5, including a TPU performance increase of up to 60% and an average CPU speed increase of 34%, allowing your Pixel to have a more responsive experience in daily use (such as web browsing and experiencing the latest AI features). The Tensor G5 is designed using TSMC’s leading 3nm process technology, which enables us to integrate more transistors into the chip, enhancing its performance and efficiency.
The new security hardware in Tensor G5 also provides better protection throughout the entire lifecycle of the Pixel (from manufacturing to the moment it is in your hands).
2. It unlocks advanced device-based AI on any phone — faster and more efficiently.
Thanks to our close collaboration with Google DeepMind, Tensor G5 is the first chip to run the latest Gemini Nano model, unlocking numerous fully device-based generative AI experiences at launch (details below!). Tensor G5 not only runs the highest quality models but does so faster and more efficiently than previous generations. In fact, with the higher performance and optimizations provided by Tensor G5 in the latest models, Gemini Nano runs 2.6 times faster and is twice as efficient in use cases like Pixel screenshot and recorder.
3. It supports proactive, practical new features for Pixel 10.
Unique Pixel features, such as Magic Cue, voice translation, call annotations with actions, and personal journals, are made possible with Tensor G5, allowing you to enjoy new proactive assistance on your device. Coupled with Gemini Nano, Tensor G5 also enhances familiar features you know and love, such as scam detection and natural voice editing in Gboard.
4. Google Tensor G5 powers the best camera ever in a Pixel.
The Pixel 10 Pro is the best smartphone photography camera, powered by the all-new image signal processor (ISP) in Tensor G5. Even in low-light conditions, our ISP helps you capture the highest quality video on Pixel every time, supporting motion blur reduction and default 10-bit video at 1080p and 4k30 resolution. Additionally, it improves true tone functionality, seamlessly capturing the different skin tones of everyone in photos or videos, allowing you to present each person’s skin tone in the most authentic way.
In addition to improvements in image and video quality, Tensor G5 supports a range of camera features, including “Add Me,” “Auto Best Shot,” and “Pro Res Zoom,” allowing the largest models running in Pixel Camera to extend zoom ranges to an astonishing 100x with stunning detail.
The camera app on Pixel 10 is the first to include C2PA content credentials. The entire process occurs on-device with the Tensor G5 and Titan M2 security chip, creating secure metadata in images that records their entire transmission process. It has achieved the highest security rating currently defined by C2PA.
5. It helps your Pixel phone last all day long.
Even with all these new features, each Pixel 10 phone has a battery life of over 30 hours. This is enough to meet your long usage needs fromthe moment you wake up until you go to sleep.
Source:
Wangcai Chip ContentCompiled fromandroidauthority
Recommended Reading
The Path of Bosch’s Chip R&D Amid Chip Shortages
How to Achieve Cabin-Driving Integration: Technical Solutions and Cabin-Driving Integration Case Sharing
Tesla’s Self-Developed AI Chip Development and Technical Effect Analysis
—end—